NVIDIA Warp is a powerful Python framework that JIT compiles functions into efficient kernel code for CPU and GPU execution, ideal for spatial computing and various applications. What sets Warp apart is its differentiable kernels, compatible with machine-learning tools like PyTorch and JAX. Installing and running Warp is straightforward using Python 3.9+, with GPU support requiring a CUDA-capable NVIDIA GPU. Developers can explore physical simulations, run examples, build the library from source, and find further resources for learning and support. Warp’s versioning closely follows Python conventions, and contributions to the project are encouraged. Don’t forget to cite Warp in your research if you utilize it.
https://github.com/NVIDIA/warp